In [2]:
#To install packages try this in condo r
#import pip
#package_name='django'
#pip.main(['install', package_name])
In [3]:
import dicom # for reading dicom files
import os # for doing directory operations
import pandas as pd
import matplotlib.pyplot as plt
import cv2
import numpy as np
import math
import glob
In [4]:
data_dir = 'D:\Data\sample_images\\'
patients = os.listdir(data_dir)
labels_df = pd.read_csv('D:\Data\\stage1_labels.csv', index_col=0)
labels_df.head()
Out[4]:
In [5]:
dcm = 'D:\Data\sample_images\\0a38e7597ca26f9374f8ea2770ba870d\\4ec5ef19b52ec06a819181e404d37038.dcm'
print('Filename: {}'.format(dcm))
dcm = dicom.read_file(dcm)
In [6]:
import matplotlib.pyplot as plt
img = dcm.pixel_array
img[img == -2000] = 0
#plt.axis('off')
#plt.imshow(img)
#plt.show()
plt.axis('off')
plt.imshow(-img) # Invert colors with -
plt.show()
# Reference couple great 1-liners from: https://www.kaggle.com/sentdex/data-science-bowl-2017/first-pass-through-data-w-3d-convnet
In [7]:
import cv2
import numpy as np
IMG_PX_SIZE = 200
for patient in patients[:1]:
label = labels_df.get_value(patient, 'cancer')
path = data_dir + patient
slices = [dicom.read_file(path + '/' + s) for s in os.listdir(path)]
slices.sort(key = lambda x: int(x.ImagePositionPatient[2]))
fig = plt.figure()
for num,each_slice in enumerate(slices[:15]):
y = fig.add_subplot(5,3,num+1)
new_img = cv2.resize(np.array(each_slice.pixel_array),(IMG_PX_SIZE,IMG_PX_SIZE))
y.imshow(new_img)
plt.show()
In [12]:
# DICOM rescale correction
data_dir = 'D:\Data\sample_images\\'
dicom_root = data_dir
def rescale_correction(s):
s.image = s.pixel_array * s.RescaleSlope + s.RescaleIntercept
# Returns a list of images for that patient_id, in ascending order of Slice Location
# The pre-processed images are stored in ".image" attribute
def load_patient(patient_id):
files = glob.glob(dicom_root + '/{}/*.dcm'.format(patient_id))
slices = []
for f in files:
dcm = dicom.read_file(f)
rescale_correction(dcm)
# TODO: spacing eq.
slices.append(dcm)
slices = sorted(slices, key=lambda x: x.SliceLocation)
return slices
In [ ]:
In [13]:
# Load a patient
for patient_no in patients:
pat = load_patient(patient_no)
print(patient_no)
img = pat[ int(len(pat)/2) ].image.copy()
# threshold HU > -300
img[img>-300] = 255
img[img<-300] = 0
img = np.uint8(img)
# find surrounding torso from the threshold and make a mask
im2, contours, _ = cv2.findContours(img,cv2.RETR_LIST,cv2.CHAIN_APPROX_SIMPLE)
largest_contour = max(contours, key=cv2.contourArea)
mask = np.zeros(img.shape, np.uint8)
cv2.fillPoly(mask, [largest_contour], 255)
# apply mask to threshold image to remove outside. this is our new mask
img = ~img
img[(mask == 0)] = 0 # <-- Larger than threshold value
# apply closing to the mask
kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE,(15,15))
img = cv2.morphologyEx(img, cv2.MORPH_OPEN, kernel) # <- to remove speckles...
img = cv2.morphologyEx(img, cv2.MORPH_DILATE, kernel)
img = cv2.morphologyEx(img, cv2.MORPH_DILATE, kernel)
img = cv2.morphologyEx(img, cv2.MORPH_CLOSE, kernel)
img = cv2.morphologyEx(img, cv2.MORPH_CLOSE, kernel)
img = cv2.morphologyEx(img, cv2.MORPH_ERODE, kernel)
img = cv2.morphologyEx(img, cv2.MORPH_ERODE, kernel)
# apply mask to image
img2 = pat[ int(len(pat)/2) ].image.copy()
img2[(img == 0)] = -2000 # <-- Larger than threshold value
# closing
#sorted_contours = sorted(contours, key=cv2.contourArea, reverse=True)
#largest_contour = max(contours, key=cv2.contourArea)
#rgb = cv2.cvtColor(img,cv2.COLOR_GRAY2RGB)
#aaa = np.concatenate( sorted_contours[1:3] )
#cv2.drawContours(rgb, [cv2.convexHull(aaa)], -1, (0,255,0), 3)
plt.figure(figsize=(12, 12))
plt.subplot(131)
plt.imshow(pat[ int(len(pat)/2) ].image)
plt.subplot(132)
plt.imshow(img)
plt.subplot(133)
plt.imshow(img2)
plt.show()
#Reference : https://www.kaggle.com/armamut/data-science-bowl-2017/getting-the-lungs-right
In [22]:
def load_patient(patient_id):
files = glob.glob(dicom_root + '/{}/*.dcm'.format(patient_id))
slices = []
for f in files:
dcm = dicom.read_file(f)
rescale_correction(dcm)
# TODO: spacing eq.
slices.append(dcm)
slices = sorted(slices, key=lambda x: x.SliceLocation)
return slices
In [23]:
import matplotlib.pyplot as plt
import seaborn as sns
import os
import glob
#import ffmpeg
%matplotlib inline
p = sns.color_palette()
import glob
def get_slice_location(dcm):
return float(dcm[0x0020, 0x1041].value)
# Returns a list of images for that patient_id, in ascending order of Slice Location
def load_patient(patient_id):
files = glob.glob('D:/Data/sample_images/{}/*.dcm'.format(patient_id))
imgs = {}
for f in files:
dcm = dicom.read_file(f)
img = dcm.pixel_array
img[img == -2000] = 0
sl = get_slice_location(dcm)
imgs[sl] = img
# Not a very elegant way to do this
sorted_imgs = [x[1] for x in sorted(imgs.items(), key=lambda x: x[0])]
return sorted_imgs
In [24]:
pat = load_patient('0acbebb8d463b4b9ca88cf38431aac69')
In [27]:
import time
def animate(pat, gifname):
# Based on @Zombie's code
fig = plt.figure()
anim = plt.imshow(pat[0], cmap=plt.cm.bone)
def update(i):
anim.set_array(pat[i])
return anim,
#a = animation.FuncAnimation(fig, update, frames=range(len(pat)), interval=50, blit=True)
#a.save(gifname, writer='ffmpeg')
#a.save('gifname', writer="ffmpeg")
print(time.clock())
plt.show()
In [28]:
animate(pat, 'test.gif')
In [ ]:
In [ ]: